Voting and Stacking-Based Ensemble methods to detect Bengali Spam SMS with proposal for comprehensive dataset

Unwanted SMS messages have become a significant nuisance in Bangladesh, making spam SMS a major problem. Detecting spam SMS in Bengali is particularly challenging due to the lack of datasets and focus in this area. This paper addresses this gap by proposing an ensemble method for detecting Bengali spam SMS using machine learning. We developed a comprehensive dataset with 5750 SMS texts, employing a user data collection method consisting of Bengali SMS messages, including legitimate and spam samples. We evaluated ten different machine learning algorithms on this dataset and evaluated the models. Our approach combines Multinomial Naıve Bayes and XGBoost algorithms utilizing voting and stacking ensemble methods. Our proposed model achieved a testing accuracy rate of 98.61%. This metric demonstrates the effectiveness of our ensemble approach in identifying spam SMS in Bengali. The results suggest that the combination of multiple classifiers can significantly enhance the accuracy of spam detection compared to individual classifiers.